It has long been suggested that gene and genome duplication play important roles in the evolution of organismal complexity. For example, work by Ohno proposed that two rounds of whole genome doubling (tetraploidy) occurred during the evolution of vertebrates: the extra genes permitting an increase in physiological and anatomical complexity. Several modifications of this 'two tetraploidies' hypothesis have been proposed, taking into account accumulating data, and there is wide acceptance of the basic scheme. In the past few years, however, several authors have raised doubts, citing lack of direct support or even evidence to the contrary. Here, we review the evidence for and against the occurrence of tetraploidies in early vertebrate evolution, and present a new compilation of molecular phylogenetic data for amphioxus. We argue that evidence in favour of tetraploidy, based primarily on genome and gene family analyses, is strong. Furthermore, we show that two observations used as evidence against genome duplication are in fact compatible with the hypothesis: but only if the genome doubling occurred by two closely spaced sequential rounds of autotetraploidy. We propose that early vertebrates passed through an autoautooctoploid phase in the evolution of their genomes.